A novel adaptive Independent Component Analysis algorithm

A novel adaptive Independent Component Analysis (NAICA) algorithm is proposed which can separate the mixture of super- and sub-Gaussian sources. Two novel models are proposed to estimate the probability density function of super- and sub-Gaussian sources respectively. In the framework of natural gradient, the parameters of two models are adaptively manipulated by online kurtosis learning. Applied to the mixture of images and speeches, experiments give good performance of NAICA algorithm.